DSP framework for FANN equalizer for application in stochastic wireless channels

Equalization in multipath fading environment offers the highest computational cost in a mobile receiver design. Since its inception, Artificial Neural Network (ANN) has been accepted for widespread applications in various fields of signal processing. ANN, with its ability to discriminate nonlinear decision boundaries, establish nonlinear functional relationship between input and output. Efforts have been made to apply the processing power of ANN to deal with the complexities of channel equalization. As computational complexity is a constraint observed in this learning based system, we attempt to use a Digital Signal Processor (DSP) based framework to accelerate the convergence time during training so that the system, with reduced latency, can be appropriately modified for inclusion as a major block of adaptive receivers suitable for high data rich and mobile environments. In this work, we present a TMS320C6713 DSK based implementation of Feedforward ANN (FANN) for identification and prediction of time varying mobile radio channels in offline mode. Significant reduction in time is observed during implementation compared to that obtained using conventional CPU. Results are also compared with that obtained from different data aided channel estimation schemes.